How the rankings actually work

The candidate universe

Before ranking, we have to decide who's eligible. The candidate universe is built from two sources:

• Wikidata SPARQL — we query for every entity that's an instance of a music-group subtype (band, rock band, supergroup, heavy-metal band, etc.) OR a human with a musician/singer/songwriter occupation, AND whose listed genres overlap the rock-music subtree (a transitively-resolved set of ~120 rock subgenres seeded from Wikidata's "rock music" taxonomy node).
• MusicBrainz — we query each of 37 rock-family tags (rock, alternative rock, indie, punk, hardcore punk, heavy metal, thrash, death, black, doom, glam, classic, hard, prog, post-rock, grunge, garage, blues rock, folk rock, country rock, gothic, post-punk, new wave, shoegaze, math, emo, pop punk, ska punk, metalcore, psychedelic, art rock, krautrock, noise rock, stoner, sludge, power, speed metal) and pull the artists tagged with each. About 51,000 entries in total.

The two universes are merged by name, deduplicated by Wikidata QID, and then filtered: an artist must have a Wikipedia article, at least one album in Wikidata, and rock genres must be at least 25% of their listed genres (or 50% if their list contains an explicit pop / hip-hop / R&B / dance marker like "teen pop", "electropop", "dance-pop", or "trap"). A small canonical bypass list covers a handful of artists like Prince and Steely Dan whose Wikidata genres are dominated by funk / R&B / jazz tags but whose status as rock canon is uncontroversial.

After filtering, we land at ~15,000 candidates. The top 500 of that pool becomes the algorithmic charts.

The five components of the score

Each candidate gets five normalized sub-scores in the range 0–1:

CriticScore (30% weight)

Source: Last.fm artist.getInfo, the listeners field. Why this proxies critical reception: Last.fm's scrobbling community has been logging plays since 2002 and is skewed toward serious music listeners — the same demographic that historically drives critic-list canonization. Listener count is log-normalized across the universe (heavy long tail) and rescaled to 0–1.

AggregateScore (15% weight)

Source: Last.fm playcount. Why this is a separate signal from listeners: Listeners measures breadth ("how many unique people have heard you"); playcount measures depth ("how many times people came back"). Some artists have huge breadth and shallow play depth (one-hit acts), others have the opposite (deep-cut bands with devoted but small audiences). Splitting them into two sub-scores prevents either pattern from dominating.

CommercialScore (10% weight)

Source: Wikipedia pageview velocity (12-month monthly average + 12-month total, equally weighted). Why we use this instead of Spotify popularity: Spotify's Web API is now Premium-gated for new applications, so we proxy current public interest with English-Wikipedia pageviews. Both signals are log-normalized.

InfluenceScore (30% weight)

Source: PageRank on the Wikidata P737 ("influenced by") directed graph, restricted to artists in our candidate universe. Edges are oriented from the influenced artist toward the influencer (so PR mass flows to canonical sources). About 2,200 edges across 15,000 nodes — a sparse graph by design (P737 is hand-curated by Wikidata editors).

Because the graph is sparse, a small number of canonical artists (Beatles, Dylan, Stones, Led Zeppelin, Stooges, Velvet Underground, Sex Pistols) accumulate most of the inbound mass. Mid-canon acts (Metallica, Radiohead, R.E.M., U2) get less because fewer Wikidata editors have explicitly cited them as influences.

LongevityScore (15% weight)

Source: Wikidata formation date (P571) and dissolution date (P576), or for solo artists, career start (first release year) and career end (death year, or "present"). Calculated as min(1, max(0.4, years / 15)). The 0.4 floor and the 15-year full-credit threshold prevent the formula from over-penalizing short but culturally seismic careers (the Beatles' 10 years, Led Zeppelin's 12 years, Nirvana's 7 years).

The formula

Each candidate's final score is:

FINAL_SCORE = 0.30 × CriticScore
            + 0.15 × AggregateScore
            + 0.10 × CommercialScore
            + 0.30 × InfluenceScore
            + 0.15 × LongevityScore

if formationYear < 1980:
    FINAL_SCORE *= 1.08    (canon-favoring era correction)

The era correction is a small thumb on the scale toward foundational rock, because Last.fm and Wikipedia pageview signals both structurally favor post-2000 acts (more recent listeners, more current search interest). Without the correction, Linkin Park and Coldplay rank above Led Zeppelin and Black Sabbath; with it, the canon roughly aligns.

Validation

After every pipeline run, we check that 38 hand-selected canonical artists land at or above their expected rank. The gate hard-fails if any miss:

• The Beatles, The Rolling Stones, Bob Dylan must be in the top 5
• Led Zeppelin, Pink Floyd, David Bowie, Jimi Hendrix in the top 15
• Black Sabbath, Queen, AC/DC, The Who, Nirvana, Radiohead, Metallica in the top 20
• The Velvet Underground, The Clash, Sex Pistols, Patti Smith, Iggy Pop, Lou Reed, Joni Mitchell, Eric Clapton in the top 50

The canonical set is sourced from Rolling Stone's "100 Greatest Artists", the top tier of the Rolling Stone "500 Greatest Albums", Rate Your Music's all-time chart, and the first-wave Rock and Roll Hall of Fame inductees. It's a check on the algorithm — not a list we produce.

Known biases and limitations

• Anglosphere bias. Last.fm and English Wikipedia both skew heavily toward English-speaking audiences. Non-Anglo bands (Spanish, French, Japanese, Australian rock) typically rank lower than their critical reputation outside our charts would suggest.
• Streaming-era recency bias. Even with the era multiplier, artists with active 2010s+ Last.fm and Wikipedia presence rank slightly higher than equivalent acts whose peak ended pre-2002 (Last.fm's launch).
• Cult-following blindness. Bands with small but devoted audiences (post-rock, sludge metal, hardcore punk, some prog) accumulate fewer listener counts than their critical importance would warrant. Examples: Pelican, Cult of Luna, Russian Circles, Mission of Burma all rank well below their consensus position.
• Sparse influence graph. Wikidata's P737 has rich coverage for the very top of the canon and thin coverage for mid-canon. Closing this gap would require ingesting actual critic-list data (Rolling Stone, Pitchfork) which is paywalled or behind aggressive scraping defenses.

These biases are visible in the ranking — they're not bugs to hide. The curated list at /charts/bands exists precisely to provide a hand-tuned counterweight when the algorithm's blind spots bite.

Reproducibility

Everything described here is deterministic given the same source data and weights. The pipeline runs in five stages — universe discovery, Last.fm fetch, Wikipedia pageview fetch, Wikidata P737 fetch, scoring + emit — each producing a JSON cache that the next stage consumes. Re-running with the same inputs produces the same ranking.